Assessing optimal set of implemented physical parameterization schemes in a multi-physics land surface model using genetic algorithm
Department of Environmental Science and Engineering
Since the concern of global warming has increasingly demanding reliable future predictions of the earth environment, the modelling techniques have been greatly advanced for the recent decades. However, as the models have been more sophisticated and complex, the uncertainties have inevitably increased mostly due to our insufficient knowledge of the nature. Especially for land surface modelling that generally represent the characteristics of land surface hydrology, biophysics, and bio-geochemistry, the dealing with accuracy of the land surface models is more challenging because of the very high heterogeneity of the land surface.
It is commonly accepted that there is no perfect model with satisfied representativeness of all embodied schemes in the land surface models. Every model has its own weaknesses induced by such problems as the reality of physical schemes by uncertain parameterizing methods and even structural unreality by simplified model designs. In addition, the lack of observation data worsens our incomplete understanding of nature.
In order to deal with model uncertainties, model optimizations through parameter calibrations have been essentially performed. Model optimization often includes priori parameter sensitivity analyses, by exploring only a limited number of parameters. However, this type of optimizations tends to be limited to only a few sites due to the tremendous computing resources and time. Considering that the models’ complexity is continuously increase, the understanding of interrelationships among the embedded schemes will be very important to improve the simulation accuracies.
Professor Park his colleagues at Ewha Womans University published a research paper in the Journal of Geoscientific Model Development in 2014, addressing the development of a novel model optimization system by coupling the multi-physics Noah land surface model(Noah MP) and the micro-genetic algorithm(micro-GA). Noah-MP was augmented with multiple physics options with regard to 10 different land surface processes on the basis of the original Noah land surface mode.
Micro-GA is an improved version of genetic algorithm that is a heuristic optimization method based on natural genetic variation and natural selection that pursue a cost-effective solution. This system basically performs scheme-based model optimization in a very effective way.
Hong et al. (2014) successfully applied the coupling system to the optimizations of multiple land surface variable simulations in terms of surface water balance (evapotranspiration and runoff) over the Han River basin in Korea, showing outstanding speeds in searching for the optimal scheme combination. The authors found that the two targeted variables were conflicting each other in terms of the accuracy of the surface water partitioning. They demonstrated, however, that the coupling system was also useful to draw reasonable outputs reducing the conflict problem.
In addition, their study showed a potential advantage of the coupling system to model diagnosis. That is, the natural selection mechanism through the micro-GA’s evolutionary process of generations (multiple sets of scheme combinations) provides information on scheme sensitivity and interrelationship that is useful to build a valuable base for further calibrations and improvements. This information is beneficial especially when the optimization is performed for more than two variables. Considering that the recent trend of increased model complexity with the consequent increase of uncertain parameters may require more efforts and time for parameter optimization and calibration, the approach such as the coupling system provides more insightful understanding of the implemented physical schemes and their interrelationship that are essential for more effective model optimization.
Figure. 2. Evolution of generations during the process of the MP-MGA optimization for (a) ET and (b) runoff. The red and blue lines indicate progresses of the ET and runoff averages, respectively, showing opposite evolution tracks each other.
Figure 3. Percentage of selections of each scheme to the total number of simulations from micro-GA during the entire optimization processes for (a) ET and (b) runoff. Each color indicates the available options in each scheme category. These analyses can provide information about what selection in each scheme category is most advantageous and hence the best optimized scheme combination.
Figure 4. (a) Evolution of generations during the process of the MP-MGA optimization for the WB estimation as in Fig. 3, and (b) percentage of selections of each scheme to the total number of simulations during the optimization process. This figure shows how the coupling system solves the conflict problem between the two targeted variables.
Assessing optimal set of implemented physical parameterization schemes in a multi-physics land surface model using genetic algorithm.
Geoscientific Model Development, Volume 7(5): 2517-2529. (2014)